The world’s fastest AI supercomputer started, 40 billion billion floating point operations per second, is stitching together the largest 3D map of the universe

See you soon!

An epic moment of computer exploration of the universe! Perlmutte, touted as the world’s fastest artificial intelligence workload supercomputer, was recently announced to be turned on. With 6,144 Nvidia A100 tensor core graphics processors, this new supercomputer will be responsible for stitching together the largest 3D map of the visible universe ever created. And, it promises to lift the dark cloud of the physics sky – dark energy.

Is the universe expanding? Yes, it is! And the “culprit” of the expanding universe is dark energy.

As the most mysterious substance in the universe, it cannot be seen or touched, and in order to capture it, human beings have established many experiments on Earth, but with little success.

The world's fastest AI supercomputer started, 40 billion billion floating point operations per second, is stitching together the largest 3D map of the universe

But today, a supercomputer with powerful AI performance may be able to help us.

Recently, NVIDIA and the National Energy Research Scientific Computing Center (NERSC) flipped a “switch” – Perlmutte, which is described as the world’s fastest supercomputer for artificial intelligence workloads.

The world's fastest AI supercomputer started, 40 billion billion floating point operations per second, is stitching together the largest 3D map of the universe

The new supercomputer, named after astrophysicist Saul Perlmutter, has 6,144 NVIDIA A100 Tensor Core GPUs and will be responsible for stitching together the largest 3D map of the visible universe ever created.

Not only that, Perlmutter will also process data from the “Cosmic Camera” Dark Energy Spectrograph (DESI), a cosmic camera that can capture up to 5,000 galaxies in a single exposure.

Processing DESI’s huge amount of data to create the largest 3D map of the visible universe
So how do you put together a 3D map of the universe?

Not long ago, on May 17, DESI was launched first, beginning a five-year journey to capture the data. During the previous four-month trial period, DESI has captured the spectra of 4 million galaxies, which is more than the sum of all previous spectroscopic surveys.

The world's fastest AI supercomputer started, 40 billion billion floating point operations per second, is stitching together the largest 3D map of the universe

What Perlmutter has to do is to aggregate the data from DESI.

According to the official website, Perlmutter’s GPU captures dozens of exposures in a single night. On previous systems, preparing a year’s worth of data for release could take weeks or months, but Perlmutter will be able to accomplish the task in just a few days.

I’m very pleased with the 20x acceleration we got on the GPU in preparation. said Rollin Thomas, a data architect at NERSC who is helping researchers prepare the code for Perlmutter.

Rollin Thomas cannot hide his confidence in the performance of Perlmutter, the world’s largest A100-driven system, with more than 20 applications on track to be the first to ship with 6,159 NVIDIA A100 Tensor Core GPUs when When NERSC is officially operational, Perlmutter will deliver nearly 4 exaflops of AI performance to more than 7,000 researchers.

The world's fastest AI supercomputer started, 40 billion billion floating point operations per second, is stitching together the largest 3D map of the universe

Dark energy was discovered primarily through the work of Saul Perlmutter, a Nobel laureate in 2011.

The world's fastest AI supercomputer started, 40 billion billion floating point operations per second, is stitching together the largest 3D map of the universe

In physical cosmology, dark energy is an undetectable form of energy that fills space and increases the rate of expansion of the universe. The dark energy hypothesis is one of the most popular explanations of the observed results of the accelerated expansion of the universe today. In the standard model of the universe, dark energy accounts for 68.3% of the mass energy of the universe

In 1998, the High Redshift Supernova Search Group published observations of Type Ia supernovae, which showed that the universe was expanding at an accelerated rate. Subsequently, in 1999, the Supernova Cosmology Project confirmed the results. This work was awarded the Nobel Prize in Physics in 2011.

But unfortunately, we still do not know enough about it, and DESI’s map gives us new hope to reveal the mystery of dark energy.

Supercomputers integrate AI, HPC
There is no doubt that Perlmutter will be infinitely useful, not only for piecing together 3D maps of the universe, but also for helping to explore subatomic interactions of green energy and more.

It used to be impossible to do a full atomic simulation of a large system like a battery interface, but now scientists plan to do it with Perlmutter,” said Brandon Cook, an application performance expert at NERSC.

We know that conventional supercomputers can barely handle the mathematical operations needed to generate several atomic simulations in a few nanoseconds with programs like Quantum Espresso. But by combining their highly accurate simulations with machine learning, scientists can study more atoms in a much longer period of time.

The world's fastest AI supercomputer started, 40 billion billion floating point operations per second, is stitching together the largest 3D map of the universe

This is where the Tensor Cores in the NVIDIA A100 play their unique role. They accelerate double-precision floating-point math operations for simulation and the mixed-precision calculations required for deep learning.

Perlmutter is based on the HPE Cray Shasta platform that includes the Slingshot interconnect, a heterogeneous system with GPU acceleration nodes and CPU-only nodes. The system is being installed in two phases – Phase 1, which was recently unveiled, includes the system’s GPU-accelerated nodes and staging file system; Phase 2 will add CPU-only nodes later in 2021.

This makes Perlmutter the fastest system on the planet for mixed 16- and 32-bit precision mathematical AI use,” said Dion Harris, senior product marketing manager at NVIDIA, in a blog post today. And is by far the second phase of a system that may be even stronger at Lawrence Berkeley National Laboratory later this year.

Perlmutter’s A100 GPU uses Nvidia Tensor Core technology and direct liquid cooling. In addition, it is NERSC’s first supercomputer with an all-flash staging file system. According to NERSC, the 35 petabyte Lustre file system will move data at over 5 terabytes per second, making it the fastest storage system in its class.

The first phase of the Perlmutter installation consists of 12 GPU-accelerated cabinets that can accommodate more than 1,500 nodes. Each GPU acceleration node in Phase 1 has four A100 Tensor Core GPUs based on the NVIDIA Ampere GPU architecture and 256GB of memory. The Phase 1 system also includes non-computing nodes (NCNs), 20 user access nodes (NCN-UAN – login nodes) and service nodes. According to NERSC, some NCN-UANs can be used to deploy containerized user environments, orchestrated using Kubernetes.

The world's fastest AI supercomputer started, 40 billion billion floating point operations per second, is stitching together the largest 3D map of the universe

Phase 1 cabinets have no connecting doors and direct liquid cooling systems with blue and red lines

Phase 2 will have two AMD Milan CPUs per CPU node and 512GB of memory per node. the Phase 2 system also adds 20 login nodes and 4 large memory nodes.

The world's fastest AI supercomputer started, 40 billion billion floating point operations per second, is stitching together the largest 3D map of the universe

Support for Multiple Programming Environments to Explore the Limits of the Universe with Computational Limits
In addition to the CCE, GNU and LLVM compilers, the Perlmutter programming environment will feature the NVDIA HPC SDK (software development kit) to support multiple parallel programming models such as MPI, OpenMP, CUDA and OpenACC for C, C++ and Fortran code.

While humans are limited in their ability to explore the universe in the flesh, computers do not have this barrier.

Last October, for example, a group of astronomers from the Institute of Astronomy at the University of Hawaii at Manoa created the most comprehensive ‘astronomical imaging catalog’ to date, including stars, galaxies and quasars, with the help of AI neural networks.

The world's fastest AI supercomputer started, 40 billion billion floating point operations per second, is stitching together the largest 3D map of the universe

The system also determines distances to galaxies with an error of up to 3%. According to the University of Hawaii, the final result is “the world’s largest 3D imaging catalog of stars, galaxies and quasars”.

There is even an 80-page paper by a physicist from Microsoft proving the “analog matrix”: the universe is a self-taught computer.

The world's fastest AI supercomputer started, 40 billion billion floating point operations per second, is stitching together the largest 3D map of the universe

A self-learning system of evolutionary laws (a self-learning system of evolutionary laws)

According to the authors, the universe has also evolved a self-learning system similar to the deep learning framework.

We know that a deep learning framework is a set of building blocks, and the individual components are part of a certain model or algorithm, and you can design your own stacks of the blocks.

Therefore, can we imagine that the universe evolved the operational matrix architecture of the law, which itself evolved from an automatic teaching system that arose from the most probable minimal initial conditions?

In the paper, the authors describe several models, all of which implement ‘self-directed evolution’.

The world's fastest AI supercomputer started, 40 billion billion floating point operations per second, is stitching together the largest 3D map of the universe

A graph with 7,088 nodes and 7,304 edges, consisting of a sampling of possible futures

In life, the laws of physics are derived by our observations, so the original laws of physics will be extremely simple, but through generations, the laws have self-perpetuating and have the ability to learn and evolve.

Perhaps the universe did not begin with the Big Bang, but simply with simple interactions between particles.

The article revolves around the Restricted Boltzmann machine (RBM).

The restricted Boltzmann machine was proposed by Hinton et al. and is a generative random neural network. The concept is very abstract and is a physics-like machine learning model. However, constrained Boltzmann machines are the simplest class of deep neural network architectures.

The architecture consists of two layers of neurons. One is the visible layer (green) and the other is the hidden layer (blue).

The world's fastest AI supercomputer started, 40 billion billion floating point operations per second, is stitching together the largest 3D map of the universe

Understanding the unknown side of the universe has always been the direction of astronomers’ efforts. With AI as a ‘teammate’, will we speed up the passage to the stellar ocean of exploring the universe?

Reference.
https://blogs.nvidia.com/blog/2021/05/27/nersc-perlmutter-ai-supercomputer/

Posted by:CoinYuppie,Reprinted with attribution to:https://coinyuppie.com/the-worlds-fastest-ai-supercomputer-started-40-billion-billion-floating-point-operations-per-second-is-stitching-together-the-largest-3d-map-of-the-universe/
Coinyuppie is an open information publishing platform, all information provided is not related to the views and positions of coinyuppie, and does not constitute any investment and financial advice. Users are expected to carefully screen and prevent risks.

Like (0)
Donate Buy me a coffee Buy me a coffee
Previous 2021-05-31 00:10
Next 2021-05-31 00:18

Related articles